PAcULTAD De INGENIEREA UNIVERSIDAD DEPARTAMENTO DE SISTEMAS & INDUSTRIAL NACIONAL siscuise Learsine 200 SEMESTRE DE 2011 Assignment 2: Bayesian decision theory (I). Anibal Montero L. 162253 Delivered to: Prof. Fabio A. Gonzalez. 1. Download the dataset from the course website. The dataset is a text file whit a number of data samples, one per line. Each line has the structure: xiyiGi_, where (xi.y,) € R? and G; € (0.1.2). ‘The following scatter plot, density plot and levels contours plot has been developed for represent the provided data set: PicURE 1, 2, & plots of the dataset.2. Use a portion of the dataset (80% of the samples) to estimate the parameters of a bivariate Gaussian distribution for each class. each class are: Considering that the unbiased estimators of the normal distribution g, leks ee, Ae k=012 5, KK = Xp) a a, k=01,2 We had the following estimations from the selected portion of the dataset: ES ese 5.121) 5.964) 3.618! 1481 —0.807 ees ost) 5 0.807 1128] * * lose 1.519. at 3. Write a program that calculates the discriminant function for each class, taking into account the possibility of rejection with a cost 2 and cost 1 for misclassification ({Alp04] Eq. (3.10)). 4. Draw the discriminant fimetions showing the boundary for each class and, implicitly the rejection area. Considering the definition of the expected risk: 0; i=k REX) =DiAgPG), Aw=jdi tek A; overwise In our case, this funetion is defined as follows RG dx) = Live PGI; k= 01,2; RCreject x) =A Then: RCL) = 1- PCI) a1 - FEC Rub) = 1 — 5 eeeicp PCD FICK) RGD 3LifleoPcaFICK) REx) SiFelco Where: exp (38K y) SEE-H) B 2nlsyl u FRIG) = ( ‘The optimal choose rule is to choose Cy if R(Gelx) = min{ R(GiIx)} and RC lx) 1-A. reject if 1-A> POW. And given the decision rule, we proceed to work whit this equations, we build the following ‘graphs of expected risk for different values of A: : y a] FIGURE 45,6: decision regions whit A= 02, 043, 05. Classify the rest of the dataset that was not used for training, using a classifier based on the discriminant functions. Evaluate the results, ‘The following table shows the sum of penalizations of the expected risk for different values of 4 in each set of data, this shows theimportance of considering the risk values according to the stage where thisis, A_training rest 2_training rest 005204 BAB Of BGT o1 388 12] fou 199 68 015 546 Md 05° BT 02 582 ua] foss 21s 7 02 a7 10a} ow 2 7 03 304 04 or 7 0.35 30.55 7.6 os 7